Skip to main content
Sets how long an agent acts on events by itself — CI results, pull-request comments and reviews, and merge conflicts in its repos. The window is measured from the last human interaction, or from the moment autopilot was last turned back on, whichever is later. Before either, it runs from the moment the agent started. Only a person restarts the clock: a follow-up you send, a PR comment you write, a review you leave. Bot comments, CI results, and the agent’s own retries do not. When the window lapses, the agent turns all three settings off and leaves them off. A later human message moves the clock but does not turn them back on — turning any of them back on is explicit, and grants a fresh full window from that moment. Nothing is terminated and no running turn is interrupted: the agent stays reachable, and still responds to anything you send it directly.

Synopsis

On a VM the slug may be omitted, which targets the calling agent itself.

Arguments

The new value takes effect on the agent’s next event, and applies to the window already running — it is not a fresh window.

Output

The timeout in force and the moment it expires are both in murmur status --json, as autopilotTimeout and autopilotDeadline, and on the agent’s dashboard page. autopilotDeadline is unset when no timeout is in force, and once the settings are already off.

Examples

Bound an agent to three days of autonomy

If nobody messages fix-auth, comments on its PR, or reviews it for 72 hours, the agent stops reacting to CI results and PR comments, sets every repo’s conflict resolution to NONE, and waits.

Let an agent run unattended indefinitely

Bound the calling agent from its own VM

Read the deadline back

Turning autopilot back on

The three settings the timeout governs are turned back on separately, and turning any one of them back on starts a fresh window:

Errors